Wind speed interval prediction method, model training method, device, equipment and medium
By using multiple gated convolutional layers and fully connected layers in the wind speed interval prediction method, the problems of low wind speed interval prediction efficiency and poor coverage in the prior art are solved, and more efficient and accurate short-term wind speed interval prediction is achieved.
Patent Information
- Application Number
- CN202411925133.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-05-23
AI Technical Summary
In the prior art, there are problems such as long prediction time, low efficiency, and poor coverage of wind speed intervals in the prediction of wind speed intervals.
A training method of short-term wind speed interval prediction model is adopted. By obtaining the original wind speed data, multiple gated convolutional layers are input, multi-scale feature extraction is performed, output is connected and inputted to the fully connected layer, the upper and lower limits of the predicted wind speed interval are obtained, and training and optimization is performed through preset loss function.
This greatly reduces the time required to train and optimize the data, improves prediction accuracy and wind speed interval coverage, and shortens prediction time.
Smart Images
Figure CN120030096A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to a wind speed range prediction method, model training method, device, equipment and medium. Background Art
[0002] As a clean and renewable energy source, wind power has shown the potential to gradually replace traditional power generation. Wind power is characterized by high intermittency and volatility, which brings profound challenges to the large-scale integration of wind power into the power system. Accurate wind forecasting can ease the task of balancing power supply and demand at the lowest cost, so there is a need to improve wind power forecasting.
[0003] At present, most of the wind speed forecasts are converted into wind power, rather than directly predicting wind power, that is, predicting the wind speed range. The intermittent and random nature of wind brings many difficulties to the prediction. In existing research, the prediction of wind speed range mainly has the problems of long prediction time, low efficiency and poor coverage of wind speed range. Summary of the invention
[0004] In order to solve the above technical problems, the present disclosure provides a wind speed interval prediction method, a model training method, an apparatus, a device and a medium.
[0005] In a first aspect, an embodiment of the present disclosure provides a training method for a short-term wind speed interval prediction model, comprising:
[0006] Acquire a training sample; the training sample includes original wind speed data;
[0007] Inputting the original wind speed data into a plurality of gated convolutional layers to obtain a plurality of gated convolutional layer outputs; the plurality of gated convolutional layers have different window sizes;
[0008] Connecting the outputs of the multiple gated convolutional layers and inputting them into a fully connected layer for processing to obtain a first linear activation output and a second linear activation output; the first linear activation output and the second linear activation output respectively represent the upper limit and the lower limit of the predicted wind speed range;
[0009] The first linear activation output, the second linear activation output and the true value in the original wind speed data are trained and optimized through a preset loss function.
[0010] In a second aspect, an embodiment of the present disclosure provides a wind speed interval prediction method, comprising:
[0011] Get wind speed data;
[0012] The wind speed data is input into a pre-trained prediction model to obtain a predicted wind speed range; the prediction model is trained using the method described in the first aspect.
[0013] In a third aspect, the present disclosure provides a method and apparatus for training a short-term wind speed interval prediction model, including:
[0014] An acquisition module, used for acquiring training samples; the training samples include original wind speed data;
[0015] A gated convolution module, used for inputting the original wind speed data into a plurality of gated convolution layers to obtain a plurality of gated convolution layer outputs; the plurality of gated convolution layers have different window sizes;
[0016] An activation module, used for connecting the outputs of the multiple gated convolutional layers and inputting them into a fully connected layer for processing to obtain a first linear activation output and a second linear activation output; the first linear activation output and the second linear activation output respectively represent an upper limit and a lower limit of a predicted wind speed interval;
[0017] A training module is used to train and optimize the first linear activation output, the second linear activation output and the true value in the original wind speed data through a preset loss function.
[0018] In a fourth aspect, an embodiment of the present disclosure provides an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the above method.
[0019] In a fifth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program implements the above method when executed by a processor.
[0020] Compared with the prior art, the technical solution provided by the disclosed embodiment has the following advantages: multiple gated convolutional layer outputs are obtained by inputting the original wind speed data into multiple gated convolutional layers with different window sizes, and the multiple gated convolutional layer outputs are connected and input into the fully connected layer for processing to obtain a first linear activation output and a second linear activation output, and the first linear activation output and the second linear activation output and the true value in the original wind speed data are trained and optimized through a preset loss function, thereby being applied to short-term wind speed interval prediction, and simultaneously using convolution operations and recursive operations, combining the advantages of recursive operations being good at combining time information and convolution operations having extremely high parallelism, thereby greatly reducing the time required for training and optimizing data, while retaining the contextual nature of the recursive neural network, improving the prediction accuracy and wind speed interval coverage, and shortening the prediction time. BRIEF DESCRIPTION OF THE DRAWINGS
[0021] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0022] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.
[0023] Figure 1 A schematic diagram of a flow chart of a training method for a short-term wind speed interval prediction model provided by an embodiment of the present disclosure;
[0024] Figure 2 A schematic diagram of a gated convolutional layer provided in an embodiment of the present disclosure;
[0025] Figure 3 A schematic diagram of a short-term wind speed interval prediction framework based on mass-driven loss and gated convolution provided by an embodiment of the present disclosure;
[0026] Figure 4 A comparison chart of a predicted wind speed range and original data provided by an embodiment of the present disclosure;
[0027] Figure 5 A schematic diagram of the structure of a wind speed interval prediction device provided in an embodiment of the present disclosure. DETAILED DESCRIPTION
[0028] In order to more clearly understand the above-mentioned objectives, features and advantages of the present disclosure, the scheme of the present disclosure will be further described below. It should be noted that the embodiments of the present disclosure and the features in the embodiments can be combined with each other without conflict.
[0029] In the following description, many specific details are set forth to facilitate a full understanding of the present disclosure, but the present disclosure may also be implemented in other ways different from those described herein; it is obvious that the embodiments in the specification are only part of the embodiments of the present disclosure, rather than all of the embodiments.
[0030] Figure 1 A flow chart of a training method for a short-term wind speed interval prediction model provided in an embodiment of the present disclosure. The method provided in an embodiment of the present disclosure can be executed by a training device for a short-term wind speed interval prediction model, which can be implemented in software and / or hardware and can be integrated on any electronic device with computing capabilities.
[0031] like Figure 1 As shown, the training method of the short-term wind speed interval prediction model provided by the embodiment of the present disclosure may include:
[0032] Step 101, obtaining training samples; the training samples include original wind speed data.
[0033] In this embodiment, the original wind speed data includes a wind speed value sequence, and the original wind speed data is expressed as (x t-n-1 ,…,x t-1 ,x t ), the training samples of the prediction model are constructed through the original wind speed data.
[0034] Step 102: input the original wind speed data into a plurality of gated convolutional layers to obtain a plurality of gated convolutional layer outputs.
[0035] Among them, multiple gated convolutional layers have different window sizes.
[0036] In one embodiment of the present disclosure, the original wind speed data is converted into a form of batch size, sequence length, and number of features, and then the converted input form is passed to a gated convolutional layer with different window sizes for multi-scale feature extraction.
[0037] In this embodiment, the original wind speed data is expressed as (x t-n-1 ,…,x t-1 ,x t ), convert the original wind speed data into the form of (B,n,p), where B is the batch size, n is the sequence length, and p is the number of features.
[0038] As an example, multiple gated convolutional layers use a first gated convolutional layer and a second gated convolutional layer with different window sizes. After obtaining the original wind speed data, the converted original wind speed data is respectively input into the first gated convolutional layer and the second gated convolutional layer for processing.
[0039] The gated convolutional layer is described below.
[0040] In this embodiment, each gated convolution layer performs three different gated convolutions at each position t of the input sequence, and the gated convolution layer includes a first gated convolution, a second gated convolution, and a third gated convolution, wherein the first gated convolution, the second gated convolution, and the third gated convolution are respectively expressed as follows:
[0041] z n =tanh(F n )=tanh(W z *X)
[0042] f n =sigmoid(F n )=sigmoid(W f *X)
[0043] o n =sigmoid(F n )=sigmoid(W i *X)
[0044] Among them, F n represents the value in the feature map at position n, W z represents the weight of the convolution filter at position z, W f represents the weight of the convolution filter at position f, W o represents the weight of the convolution filter at position o, and X represents the input sequence value. The first gated convolution is used to decide what information to get from the current input x t Add to current state C t , the second gated convolution is used to decide from the previous state C t-1 How much to remember and how much to get from the current input z t The third gated convolution is used to decide what to keep from the current state C. t Passed to the output.
[0045] Optionally, the original wind speed data is input into a plurality of gated convolutional layers to obtain a plurality of gated convolutional layer outputs, including:
[0046] For any gated convolution layer, the current state is updated by the first gated convolution and the second gated convolution, wherein the state of the current time step is updated by combining the information from the previous state, for example, the current state is updated using the following formula:
[0047] C n =f n ·C n-1 +(1-f n )·z n
[0048] Among them, C n is the current state, C n-1 is the previous state, z n represents the first gated convolution, f n represents the second gated convolution;
[0049] The gated convolution layer output is obtained by applying the third gated convolution to the hyperbolic tangent of the current state; the gated convolution layer output is calculated using the following formula:
[0050] H n =o n tanh(C n )
[0051] Among them, H n is the output of the gated convolutional layer, o n Represents the third gated convolution.
[0052] Reference Figure 2 , Figure 2 shows a schematic diagram of a gated convolutional layer, Figure 2The three gated convolutions can run in parallel on the input length. The output layer part represents the above process of determining the output based on the state and gated convolution. The output layer part runs sequentially on the input length.
[0053] Step 103, connect the outputs of multiple gated convolutional layers and input them into a fully connected layer for processing to obtain a first linear activation output and a second linear activation output.
[0054] In this embodiment, each gated convolution layer output is connected to form a multi-scale feature extractor, and the connected output is passed to the fully connected layer to generate two linear activation outputs. The first linear activation output and the second linear activation output represent the upper limit and the lower limit of the predicted wind speed range, respectively.
[0055] Step 104: training and optimizing the first linear activation output, the second linear activation output, and the true value in the original wind speed data by using a preset loss function.
[0056] In this embodiment, the loss is calculated for the first linear activation output, the second linear activation output and the true value in the original wind speed data by using a preset loss function to iteratively optimize the prediction model.
[0057] In one embodiment of the present disclosure, the preset loss function QD imp It is expressed as follows:
[0058]
[0059] Among them, MPIW soft is the average prediction interval width, λ is the multiplier variable introduced in the Lagrangian method, α represents the required confidence level of the generated wind speed interval, and n is the total number of observation data;
[0060]
[0061] Where c is the total number of observation data captured in the wind speed interval, is the upper limit of the predicted data points, is the lower limit of the predicted data point, k soft is the total number of wind speed intervals that can capture data points;
[0062]
[0063] Where s is the softening coefficient, ⊙ is the element product,
[0064]
[0065] PICP soft is the covering index, is the calibration function.
[0066] 6. The method of claim 5, wherein the calibration function is expressed as follows:
[0067]
[0068] Among them, y i is the true value.
[0069] In this embodiment, a quality-driven loss function that introduces a calibration function is used to improve the coverage of predicted wind speed intervals, especially to improve the multi-level prediction performance, and to generate higher quality wind speed intervals.
[0070] Reference Figure 3 , Figure 3 A schematic diagram of a short-term wind speed interval prediction framework based on mass-driven loss and gated convolution is shown in Figure 3 As shown in the figure, the framework includes two gated convolutional layers with different window sizes. The wind speed sequence is input into the gated convolutional layer to obtain the gated convolutional layer output, and then the gated convolutional layer output is connected to form a multi-scale feature representation. The multi-scale feature representation is input into the fully connected layer to obtain two linear activation outputs. The model training optimization is performed based on the above loss function, two linear activation outputs and the true value of the data. Figure 4 , Figure 4 A comparison chart of the predicted wind speed range and the original data is shown.
[0071] According to the technical solution of the embodiment of the present disclosure, the original wind speed data is input into multiple gated convolutional layers with different window sizes to obtain multiple gated convolutional layer outputs, and the multiple gated convolutional layer outputs are connected and input into the fully connected layer for processing to obtain the first linear activation output and the second linear activation output. The first linear activation output and the second linear activation output and the true value in the original wind speed data are trained and optimized through a preset loss function. Thus, it is applied to short-term wind speed interval prediction, and convolution operations and recursive operations are used at the same time, combining the advantages of recursive operations being good at combining time information and convolution operations having extremely high parallelism, which greatly reduces the time required for training and optimizing data, while retaining the contextual nature of the recursive neural network, improving the prediction accuracy and wind speed interval coverage, and shortening the prediction time. In addition, a calibration function is introduced into the traditional quality-driven loss function, and combined with gated convolution, which further improves the prediction accuracy and shortens the prediction time.
[0072] The following is an explanation based on actual data sets.
[0073] Specific data preparation:
[0074] Wind speed datasets of offshore wind farms and onshore wind farms are selected for testing. The datasets are divided into four parts, representing spring, summer, autumn, and winter, and training and test sets are set at the same time. After setting the data, the short-term wind speed interval prediction method based on mass-driven loss and gated convolution is used for prediction. Table 1 shows some metrics calculated based on the prediction results.
[0075] Table 1 Prediction indicators
[0076]
[0077]
[0078] In Table 1, PICP indicates the number of target values covered by the wind speed interval. The larger the PICP, the more reliable the wind speed interval. From Table 1, it can be seen that the wind speed intervals predicted by the two wind farms in four seasons have a high coverage rate and can be maintained above 90%; PINRW indicates the width of the wind speed interval. PINRW and PICP are contradictory, so a comprehensive evaluation is required. From Table 1, it can be seen that a smaller wind speed interval width can be basically guaranteed under a higher coverage rate, indicating that the predicted wind speed interval has a higher quality; INAD indicates the degree to which the data points not in the wind speed interval deviate from the wind speed interval. From Table 1, it can be seen that the values of INAD are relatively small, indicating that the predicted wind speed interval has a good rationality; the training time is also greatly reduced. In order to more clearly reflect these indicators, the indicators of the eight data sets are averaged, as shown in Table 2.
[0079] Table 2 Average values of prediction indicators
[0080]
[0081] It can be seen from Table 2 that the wind speed range predicted by the present invention not only has high quality, but also maintains relative rationality, while also greatly reducing the training time.
[0082] Based on the above embodiments, the present disclosure also proposes a short-term wind speed interval prediction method, which includes the following steps:
[0083] Get wind speed data.
[0084] The wind speed data is input into the pre-trained prediction model to obtain the predicted wind speed range.
[0085] The prediction model is obtained by using the training method described in the above embodiment.
[0086] In the disclosed embodiment, when predicting short-term wind speed intervals, convolution operations and recursive operations are simultaneously used, combining the advantages of recursive operations being good at combining time information and convolution operations having extremely high parallelism, thereby greatly reducing the time required for training and optimizing data, while retaining the contextual nature of the recursive neural network, improving prediction accuracy, and shortening prediction time.
[0087] Figure 5 A schematic diagram of a short-term wind speed interval prediction model training device provided by an embodiment of the present disclosure is shown in FIG. Figure 5 As shown, the training device of the short-term wind speed interval prediction model includes: an acquisition module 51, a gated convolution module 52, an activation module 53, and a training module 54.
[0088] An acquisition module 51 is used to acquire training samples; the training samples include original wind speed data;
[0089] A gated convolution module 52, used for inputting the original wind speed data into a plurality of gated convolution layers to obtain a plurality of gated convolution layer outputs; the plurality of gated convolution layers have different window sizes;
[0090] An activation module 53 is used to connect the outputs of multiple gated convolutional layers and input them into a fully connected layer for processing to obtain a first linear activation output and a second linear activation output; the first linear activation output and the second linear activation output respectively represent the upper limit and the lower limit of the predicted wind speed range;
[0091] The training module 54 is used to perform training optimization on the first linear activation output, the second linear activation output and the true value in the original wind speed data by using a preset loss function.
[0092] In one embodiment of the present disclosure, the gated convolution layer includes a first gated convolution, a second gated convolution, and a third gated convolution, and the gated convolution module 52 is specifically used for:
[0093] For any gated convolution layer, the current state is updated by the first gated convolution and the second gated convolution; wherein, the current state is updated using the following formula:
[0094] C n =f n ·C n-1 +(1-f n )·z n
[0095] Among them, C n is the current state, C n-1 is the previous state, z n represents the first gated convolution, f n represents the second gated convolution;
[0096] The gated convolutional layer output is obtained by applying a third gated convolution to the hyperbolic tangent of the current state. The gated convolutional layer output is calculated using the following formula:
[0097] H n = o n ·tanh(C n )
[0098] where H n is the gated convolutional layer output, and o n represents the third gated convolution.
[0099] In one embodiment of the present disclosure, each gated convolutional layer performs a first gated convolution, a second gated convolution, and a third gated convolution at each position of the input sequence. The first gated convolution, the second gated convolution, and the third gated convolution are expressed as follows:
[0100] z n = tanh(F n ) = tanh(W z * X)
[0101] f n = sigmoid(F n ) = sigmoid(W f * X)
[0102] o n = sigmoid(F n ) = sigmoid(W o * X)
[0103] where F n represents the value in the feature map at position n, W z represents the weight of the convolutional filter at position z, W f represents the weight of the convolutional filter at position f, W o represents the weight of the convolutional filter at position o, and X represents the input sequence value.
[0104] In one embodiment of the present disclosure, multiple gated convolutional layers adopt a first gated convolutional layer and a second gated convolutional layer with different window sizes. The gated convolution module 52 is specifically configured to:
[0105] Convert the original wind speed data into the form of batch size, sequence length, and number of features;
[0106] Input the converted original wind speed data into the first gated convolutional layer and the second gated convolutional layer for processing respectively.
[0107] In one embodiment of the present disclosure, the preset loss function QD imp is expressed as follows:
[0108]
[0109] Among them, MPIW soft is the average prediction interval width, λ is the multiplier variable introduced in the Lagrangian method, α represents the required confidence level of the generated wind speed interval, and n is the total number of observation data;
[0110]
[0111] Where c is the total number of observation data captured in the wind speed interval, is the upper limit of the predicted data points, is the lower limit of the predicted data point, k soft is the total number of wind speed intervals that can capture data points;
[0112]
[0113] Where s is the softening coefficient, ⊙ is the element product,
[0114]
[0115] PICP soft is the covering index, is the calibration function.
[0116] In one embodiment of the present disclosure, the calibration function is expressed as follows:
[0117]
[0118] Among them, y i is the true value.
[0119] The disclosed embodiment also provides a short-term wind speed interval prediction device, and the short-term wind speed interval prediction method device includes: a collection module and a prediction module.
[0120] Among them, the acquisition module is used to obtain wind speed data;
[0121] The prediction module is used to input wind speed data into a pre-trained prediction model to obtain a predicted wind speed range; the prediction model is trained using the method of the aforementioned embodiment.
[0122] The device provided in the embodiment of the present disclosure can execute any method provided in the embodiment of the present disclosure, and has the corresponding functional modules and beneficial effects of the execution method. The contents not described in detail in the embodiment of the device of the present disclosure can refer to the description in any method embodiment of the present disclosure.
[0123] The present disclosure also provides an electronic device, which includes one or more processors and a memory. The processor may be a central processing unit (CPU) or other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device to perform desired functions. The memory may include one or more computer program products, and the computer program product may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache), etc. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on a computer-readable storage medium, and the processor may run the program instructions to implement the method of the above embodiment of the present disclosure and / or other desired functions. Various contents such as input signals, signal components, noise components, etc. may also be stored in the computer-readable storage medium.
[0124] In one example, the electronic device may further include: an input device and an output device, and these components are interconnected through a bus system and / or other forms of connection mechanisms. In addition, the input device may also include, for example, a keyboard, a mouse, etc. The output device may output various information to the outside, including determined distance information, direction information, etc. The output device may include, for example, a display, a speaker, a printer, and a communication network and a remote output device connected thereto, etc. In addition, according to the specific application, the electronic device may also include any other appropriate components such as a bus, an input / output interface, etc.
[0125] In addition to the above methods and devices, the embodiments of the present disclosure may also be a computer program product, which includes computer program instructions. When the computer program instructions are executed by a processor, the processor executes any method provided by the embodiments of the present disclosure.
[0126] The computer program product may be written in any combination of one or more programming languages to write program code for performing the operations of the disclosed embodiments, including object-oriented programming languages such as Java, C++, etc., and conventional procedural programming languages such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.
[0127] In addition, an embodiment of the present disclosure may also be a computer-readable storage medium storing computer program instructions, which, when run by a processor, cause the processor to execute any method provided by the embodiments of the present disclosure.
[0128] The computer-readable storage medium may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may, for example, include but is not limited to an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
[0129] It should be noted that, in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover a non-exclusive inclusion, such that a process, method, article or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.
[0130] The above are only specific embodiments of the present disclosure, enabling those skilled in the art to understand or implement the present disclosure. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure will not be limited to these embodiments described herein, but rather will be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A training method for a short-term wind speed interval prediction model, characterized in that: include: Get training samples; The training samples include original wind speed data; Inputting the original wind speed data into a plurality of gated convolutional layers to obtain a plurality of gated convolutional layer outputs; The multiple gated convolutional layers have different window sizes; Connecting the outputs of the multiple gated convolutional layers and inputting them into a fully connected layer for processing to obtain a first linear activation output and a second linear activation output; The first linear activation output and the second linear activation output represent the upper limit and the lower limit of the predicted wind speed interval respectively; The first linear activation output, the second linear activation output and the true value in the original wind speed data are trained and optimized through a preset loss function.
2. The method according to claim 1, characterized in that The gated convolution layer includes a first gated convolution, a second gated convolution, and a third gated convolution. The raw wind speed data is input into a plurality of gated convolution layers to obtain a plurality of gated convolution layer outputs, including: For any gated convolution layer, the current state is updated by the first gated convolution and the second gated convolution; wherein the current state is updated by the following formula: C n =f n ·C n-1 +(1-f n )·z n Among them, C n is the current state, C n-1 is the previous state, z n represents the first gated convolution, f n represents the second gated convolution; The gated convolution layer output is obtained by applying the third gated convolution to the hyperbolic tangent of the current state; wherein the gated convolution layer output is calculated using the following formula: H n =o n ·tanh(C n ) Among them, H n is the output of the gated convolutional layer, o n Represents the third gated convolution.
3. The method according to claim 2, characterized in that Each of the gated convolutional layers performs the first gated convolution, the second gated convolution, and the third gated convolution at each position of the input sequence, wherein the first gated convolution, the second gated convolution, and the third gated convolution are expressed as follows: z n =tanh(F n )=tanh(W z *X) f n =sigmoid(F n )=sigmoid(W f *X) o n =sigmoid(F n )=sigmoid(W o *X) Among them, F n represents the value in the feature map at position n, W z represents the weight of the convolution filter at position z, W f represents the weight of the convolution filter at position f, W o represents the weight of the convolution filter at position o, and X represents the input sequence value.
4. The method according to claim 2, characterized in that The multiple gated convolutional layers use a first gated convolutional layer and a second gated convolutional layer with different window sizes, and the raw wind speed data is input into the multiple gated convolutional layers, comprising: Converting the raw wind speed data into a format with a batch size, sequence length, and number of features; The converted original wind speed data is respectively input into the first gated convolution layer and the second gated convolution layer for processing.
5. The method according to claim 1, characterized in that The preset loss function QD imp It is expressed as follows: Among them, MPIW soft is the average prediction interval width, λ is the multiplier variable introduced in the Lagrangian method, α represents the required confidence level of the generated wind speed interval, and n is the total number of observation data; Where c is the total number of observation data captured in the wind speed interval, is the upper limit of the predicted data points, is the lower limit of the predicted data point, k soft is the total number of wind speed intervals that can capture data points; Where s is the softening coefficient, ⊙ is the element product, PICP soft is the covering index, is the calibration function.
6. The method according to claim 5, characterized in that The calibration function is expressed as follows: Among them, y i is the true value.
7. A short-term wind speed interval prediction method, characterized in that: include: Get wind speed data; The wind speed data is input into a pre-trained prediction model to obtain a predicted wind speed range; the prediction model is trained using the method as claimed in claim 1.
8. A training device for a short-term wind speed interval prediction model, characterized in that: include: An acquisition module, used to acquire training samples; The training samples include original wind speed data; A gated convolution module, used for inputting the original wind speed data into a plurality of gated convolution layers to obtain a plurality of gated convolution layer outputs; the plurality of gated convolution layers have different window sizes; An activation module, used for connecting the outputs of the multiple gated convolutional layers and inputting them into a fully connected layer for processing to obtain a first linear activation output and a second linear activation output; The first linear activation output and the second linear activation output represent the upper limit and the lower limit of the predicted wind speed interval respectively; A training module is used to train and optimize the first linear activation output, the second linear activation output and the true value in the original wind speed data through a preset loss function.
9. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the method described in any one of claims 1 to 7 is implemented.